A Machine Learning Heat Flow Model of Antarctica

We established a new Geothermal Heat Flow (GHF) model for Antarctica by using a machine learning approach. GHF is substantially related to the geodynamic setting of the plates, and global geophysical and geological data sets can provide information for remote regions like Antarctica, where only sparse direct measurements exist. We applied a Gradient Boosted Regression Tree algorithm in order to build an optimal prediction model relating GHF to the observables.
Employed data sets are reviewed for their reliability and quality in polar regions and we emphasize the need for adding reasonable data to the algorithm. The validity of our approach is indicated by predictions for Australia, where an extensive database of GHF measurements exists. Our new estimated GHF map exhibits rather moderate values compared to previous models, ranging from 35 to 156 mWm^-2, and shows visible connections to the conjugate margins in Australia, Africa, and India.
Such estimates on the geothermal structure of Antarctica are for example needed for studies on ice sheet modeling. The internal thermal structure and the mass balance of the modeled Antarctic ice sheet (AIS) are significantly affected by the prescribed GHF distribution.  Applying a wide range of possible GHF maps within estimated uncertainties to ice-sheet-shelf set-ups, the influence of GHF on the modeled AIS response to a variety of climate scenarios is quantified.